Dynamic model identification for CNC machine tool feed drives from in-process signals for virtual process planning

Abstract

Virtual process planning offers advantages in terms of predicting servo and contouring errors ahead of time, and taking corrective action by modifying the program or CNC parameters. Successful application of virtual process planning requires accurate models that capture the dynamics of feed drives. Identification of such models is typically time consuming.This paper presents a pole search method used in conjunction with least squares (LS) projection technique for identifying virtual machine tool drives. Compared to earlier research published in rapid identification, and the generic system identification method (i.e., a two-stage instrumental variable method), the parameter convergence has been improved significantly. This is achieved by reducing the number of unknown variables solved during LS estimation from eight to four (corresponding to each candidate pole triplet), and by avoiding the use of noisy position measurements and their time derivatives in the LS matrix pseudo-inversion. As a result, virtual feed drive models can be constructed using only in-process gathered data, without interrupting a machine tool's production for dedicated identification tests. Effectiveness of the new method is demonstrated in simulations and experimental case studies on two different machine tools, a gear grinding machine, and a 5-axis machining center. With the new method, servo errors can be predicted during the process planning stage to within 1-2% of closeness of their actual (experimental) values.